Constituting a dataset for applying Natural Language Inference to Chinese Clinical Trials: possible approaches and challenges
Constitution d'un jeu de données pour l'Inférence en Langue Naturelle appliquée aux essais cliniques en chinois : possibles approches et défis
Abstract
Natural Language Inference is a Natural Language Understanding task aiming to determine the entailment relation between a given premise and hypothesis. Where we apply this task to clinical trials. We consider Clinical Trial Reports (CTR) as premises and want to determine their entailment relation with a given hypothesis. Currently, the only dataset available for this task focuses only on English. Therefore, we propose two approaches to build a new dataset using clinical trials in Chinese. The first approach consists of collecting original CTRs from Chinese clinical trial databases.
The second approach uses neural machine translation models to translate the NLI4CT dataset from English to Chinese. We evaluated a sample of our translations against human references and obtained a SacreBLEU score of up to 29.11 and a BertScore of 83.62 using mBART.
Origin | Files produced by the author(s) |
---|---|
licence |